Enhancing Satellite Data Integrity Through Online Learning for Memory Dump Scheduling

Managing data downlinks through memory dump scheduling in spacecraft operations is paramount to maintaining data integrity and keeping high mission performance. Traditional static scheduling methods lack the flexibility to adapt to random events and place a significant amount of manual workload on operators. Although machine learning techniques have shown promise, they generally require large datasets and high computational resources, both of which can be limited in practice. This paper proposes to optimize time offsets for memory dump using online learning techniques, specifically, leveraging follow-the-leader strategies. These lightweight sequential algorithms are based on the intuitive idea of choosing offsets with the currently best historical performance and are known to be optimal under realistic assumptions. By integrating real-time telemetry feedback and online learning, the proposed method dynamically adjusts memory dump timings to account for variations in spacecraft operations and ground station availability, reducing the probability of data loss due to memory saturation or ground station outage. The proposed algorithm is tested using data coming from live telemetry within the mission planning system of Sentinel-6A, demonstrating its effectiveness in optimizing memory dump by showing a remarkable improvement in data key performance index. The algorithm achieved an 86% reduction in data loss relative to the loss experienced in the real-world scenario.